TL;DR

Anthropic’s economics team has published an interactive model sketching three versions of the US economy in 2030. In the milder two, unemployment stays within historical norms; only in an extreme case, where AI growth exceeds anything on record, do knowledge workers face deep job and wage losses. The work is US-specific, but its central finding about who absorbs the shock applies to any services-heavy economy, the UK’s included.

Three futures

The model breaks each occupation into its component tasks, which AI can augment, automate, leave alone or add to. It rests on a technical paper by economist Anton Korinek, Chad Jones and three co-authors.

In the modest scenario, AI’s effect resembles the internet’s: real, gradual and hard to spot in the macro data. The substantial path gives AI the ability, by 2030, to do 50% of knowledge-work tasks, yet most still get done without it; growth runs at double its usual pace, pay for non-knowledge roles rises and knowledge workers’ wages stall. The extreme scenario assumes self-improving systems and rapid uptake, pushing annual growth to 15% while unemployment climbs past typical recession levels.

A consistent thread runs through all three. Workers currently take about 60 cents of every dollar produced, and that portion drops markedly in the two bigger scenarios even where average wages rise.

Anthropic also polled 10,980 Americans in August. The typical respondent’s expectations implied something near the substantial path, with output 10% higher than a no-AI baseline and unemployment around 5%. Roughly one in ten held views matching the extreme case.

Caveats the authors volunteer

Named reviewers including Daron Acemoglu and David Autor commented on the draft, though none was asked to endorse its conclusions. Some felt the extreme case reads more like a thought experiment; others thought the gentlest case too cautious given evidence already emerging. The model ignores policy responses, business cycles and demand from the data centre build-out.

Looking forward

For UK readers, the useful point is structural rather than numerical. Harm in these scenarios lands on knowledge workers first, and Britain’s economy leans heavily on exactly that kind of professional and financial services work. Anthropic has not run the exercise for Britain, and the OBR’s own AI assumptions span a wide range.

It is also worth remembering who built this. A lab whose products drive the disruption is modelling its consequences, and says the results will shape the labour-market research and policy ideas it funds.